Context: This month, AI copyright litigation went international on a large scale with German music rights collecting society suing OpenAI in Munich (November 14, 2024 ai fray article) and a lawsuit by India’s Asian News International (ANI) agency surfacing in the Delhi High Court (November 18, 2024 ai fray article). At today’s Indian court hearing it turned out that OpenAI has already ceased to use ANI’s coprighted works for training purposes (November 19, 2024 ai fray article).
What’s new: The CEO and the General Counsel of GEMA held a webinar today for the purpose of a background briefing. It was attended by various journalists. ai fray was also invited and gladly participated. The two officials were quite forthcoming about the rationale and the objectives of this litigation. They also explained the difference between a model action (which is how they categorize their case) and a class action.
Direct impact: It became clear that this case is unlikely to settle. It was noted that the case might even go up straight from the court of first instance (the Munich I Regional Court) to the top EU court, the European Court of Justice (ECJ) in Luxembourg.
Wider ramifications: GEMA stressed the fundamental differences between the U.S. and European legal frameworks, and seeks legal clarification of Europe-wide effect. While GEMA can obviously not speak for similar organizations in other countries, one could read between the lines that the likelihood of similar actions by collecting societies in other European countries is rather high. Its peers may, however, wait and see what headway GEMA makes.
CEO Dr. Tobias Holzmueller (“Holzmüller” in German) and General Counsel Dr. Kai Welp were the speakers at today’s background briefing, which was moderated by GEMA’s communications chief Ursula Goebel.
Dr. Holzmueller described the complaint as a “response to the situation” and said GEMA’s members (this case is about song lyrics) “feel like victims of theft because their works, without pay or permission, have been read, analyzed and used as a basis for the creation of AI-generated content that competes with what our members do.” But this perspective was nuanced a bit: GEMA acknowledged that many of its members have experimented with, or are actually using, AI systems, and that there is a mix of fascination and fear that this new technology instills in them. According to a poll, 70% of GEMA’s members are concerned that Generative AI has, or will have, the ability to compete with their works.
GEMA suggested the complaint was only a last resort after proposing different licensing models. Allegedly, OpenAI’s response to an initial demand letter was formalistic and devoid of substance. GEMA concluded, based on how OpenAI dealt with them, that it was impossible to conduct negotiations on equal footing, short of bringing suit.
Where GEMA would like to get is that AI providers pay for input as well as output:
- They want to collect royalties on the use of copyrighted material for training AI systems.
- They additionally seek to collect royalties on output that competes with their licensors’ works.
With a view to changes that OpenAI may make (or may already have made as we speak), GEMA ensured that certain ChatGPT outputs were properly documented. They stressed that they took precautions so as to ensure that ChatGPT was not retrieving the answers from the internet in real time, but was using its internal data.
Dr. Welp gave one specific example of song lyrics from GEMA’s portfolio that ChatGPT allegedly put out in response to unspecified prompts: Herbert Groenemeyer’s Bochum, a 40-year-old song by one of Germany’s most popular singers about the working-class charm of his industrial home town in the state of North Rhine-Westphalia. As GEMA noted, the lyrics of that song have been republished in many places, but GEMA declared an opt-out in writing and believes that European law guarantees the effect of opt-out declarations. They did not say so explicitly, but whether an opt-out declaration must be machine-readable (even though it is, as GEMA accurately noted, not feasible to attach it to each and every republication of the lyrics of a popular song) or not is presumably one of the legal questions on which they seek clarification.
An important admission speaks to GEMA’s outspokenness at today’s background briefing: a verbatim output of song lyrics by ChatGPT is apparently much less likely to happen in the case of a lesser-known song. Obviously, AI systems analyze statistical correlations, and if a song text is found in many documents, then certain sequences of words will potentially be reproduced. GEMA argues, however, that if they can prove reproduction based on a few German megahits like Bochum, ChatGPT will presumably have parsed less famous songs, too.
ai fray asked three questions (first two, and in another round a third one, as other attendees also asked repeatedly):
- Did GEMA get the impression that OpenAI was not sufficiently aware of fundamental differences between U.S. and European copyright laws?
GEMA believes there were no misconceptions on OpenAI’s part. In this context they noted that German copyright law is even stricter than EU law in this regard. GEMA assumes that OpenAI is a sophisticated litigant that knows what it’s doing. - Why did GEMA elect to sue in Munich? Did the Munich I Regional Court’s reputation as a strict patent enforcement venue play a role?
GEMA answered that it was the obvious choice as they are based there, though they confirmed that legally they could have sued anywhere in Germany. They added a second consideration: they want their case to be adjudicated as swiftly as possible, so the speed of the forum was a factor. In this context, GEMA also mentioned that this matter will likely have to be decided by the highest court, and that a preliminary reference to the ECJ (i.e., a request for clarification of questions of EU law prior to a ruling by the referring court) might happen even from the court of first instance. While GEMA did not say so, ai fray believes that GEMA may have taken note of the fact that the Munich I Regional Court has already referred two questions of intellectual property law to the ECJ, though only the highest court of an EU member state has an obligation to do so if a serious question of EU law is raised. A couple of years ago, a preliminary reference from Munich made preliminary injunctions over patents more widely available in the EU. And last month the Munich I Regional Court’s 21st Civil Chamber (one of the IP law divisions) referred a question concerning exclusive marketing rights for medications to the ECJ. - Is GEMA primarily seeking an injunction in order to force OpenAI to come to the negotiating table?
GEMA had previously explained that this is a model action over only a few of the copyrighted works in its portfolio (as opposed to a class action seeking compensation for all right holders). They want to prove that “this is not like in the United States where some questions are open for the time being.” GEMA also made it clear at today’s briefing that they do not want to prevent OpenAI from using its material, but seek to ensure what they can consider fair compensation. In response to ai fray‘s specific question about the primary remedy, GEMA acknowledged that injunctive relief is key, but noted that there are also other claims, such as the request for an accounting (information that would enable GEMA to calculate a damages claim).
One attendee wanted to know more about the possibility of lawsuits by collecting societies in other EU member states. GEMA noted that they are in a particularly strong position to pursue this litigation. For example, they have song lyrics in their portfolio, which some other collecting societies don’t. They have a mandate to enforce the right to reproduce. But they can’t announce complaints in other countries as there is only a “loose network” of such organizations. They communicate, and there appears to be significant coordination, but one will have to “see what happens.”
There is no question that this German case is going to be one of the most important AI copyright cases in the near term, especially given its potentially EU-wide ramifications. That is a market of 450 million people, and whatever happens in Europe can influence other jurisdictions in the world, at least in the form of persuasive authority. And ultimately this has a political dimension, too.